The Impact of the Emoticons Walter and d'Addario

http://ssc.sagepub.com

Social Science Computer Review

2001; 19; 324 Social Science Computer ReviewJoseph B. Walther and Kyle P. D'Addario

The Impacts of Emoticons on Message Interpretation in Computer–Mediated Communication

http://ssc.sagepub.com/cgi/content/abstract/19/3/324 The online version of this article can be found at:

Published by:

http://www.sagepublications.com

can be found at:Social Science Computer Review Additional services and information for

http://ssc.sagepub.com/cgi/alerts Email Alerts:

http://ssc.sagepub.com/subscriptions Subscriptions:

http://www.sagepub.com/journalsReprints.navReprints:

http://www.sagepub.com/journalsPermissions.navPermissions:

© 2001 SAGE Publications. All rights reserved. Not for commercial use or unauthorized distribution. at PENNSYLVANIA STATE UNIV on February 5, 2008 http://ssc.sagepub.comDownloaded from

http://ssc.sagepub.com/cgi/alerts

http://ssc.sagepub.com/subscriptions

http://www.sagepub.com/journalsReprints.nav

http://www.sagepub.com/journalsPermissions.nav


SOCIAL SCIENCE COMPUTER REVIEWWalther, D’Addario / EMOTICONS

The Impacts of Emoticonson Message Interpretation inComputer-Mediated Communication

JOSEPH B. WALTHERKYLE P. D’ADDARIO

Rensselaer Polytechnic Institute

Emoticons are graphic representations of facial expressions that many e-mail users embed in their mes-sages. These symbols are widely known and commonly recognized among computer-mediated com-munication (CMC) users, and they are described by most observers as substituting for the nonverbalcues that are missing from CMC in comparison to face-to-face communication. Their empiricalimpacts, however, are undocumented. An experiment sought to determine the effects of three commonemoticons on message interpretations. Hypotheses drawn from literature on nonverbal communicationreflect several plausible relationships between emoticons and verbal messages. The results indicatethat emoticons’ contributions were outweighed by verbal content, but a negativity effect appeared suchthat any negative message aspect—verbal or graphic—shifts message interpretation in the direction ofthe negative element.

Keywords: emoticon, smiley, computer-mediated communication, electronic mail,affect

There is little argument that e-mail is the most commonly used form of computer-mediated communication (CMC), and for many, it is becoming one of the most com-

mon forms of communication overall (Kraut, Mukhopadhyay, Szczypula, Kiesler, &Scherlis, 1999; Pew Internet & American Life Project, 2000). Some project that e-mail usemay even surpass traditional face-to-face (FtF) interaction in the next generation(Negroponte, 1995). However, how e-mail compares to other forms of communicationremains unclear in some respects. As traditional (ASCII-based) e-mail has no sound orgraphic component, some argue that the social functions normally communicated by nonver-bal cues in FtF encounters do not occur in CMC. Kiesler, Siegel, and McGuire (1984)observed that

traditional forms of communication, head nods, smiles, eye contact, distance, tone of voice, andother nonverbal behavior give speakers and listeners information they can use to regulate, mod-ify, and control exchanges. Electronic communication may be inefficient for resolvingsuch . . . problems. (p. 1125)

Echoing a similar point, Sarbaugh-Thompson and Feldman (1998) suggested that the lack ofnonverbal cues in CMC limits “the range of communication . . . ; sarcasm, for example, is notwell expressed on electronic mail” (p. 686). Due to these and other restrictions, some suggest

324

Social Science Computer Review, Vol. 19 No. 3, Fall 2001 324-347© 2001 Sage Publications



that e-mail is limited in scope and best suited for task-oriented purposes. As Rice and Love(1987) summarized the prevailing literature, “as bandwidth narrows, media allow less ‘so-cial presence’; communication is likely to be described as less friendly, emotional, or per-sonal and more serious, businesslike, or task oriented” (p. 88).

Despite these limitations, users have found ways to increase the richness of CMC andachieve socially oriented communication through it. Evidence establishes that CMC issometimes used for explicitly social purposes (McCormick & McCormick, 1992). In othercircumstances, users work within the limitations of text-based e-mail to achieve levels ofrelational communication equal to, or exceeding, parallel FtF encounters (e.g., Walther,1997; Walther & Burgoon, 1992). They do so by a variety of means, including their interpre-tation of natural language, questions and disclosures (Tidwell & Walther, 2000), attention tochronemic cues (business hours vs. after hours, swift reply vs. slow reply; Walther &Tidwell, 1995), and other devices (see Carey, 1980).

Another way in which e-mail users may imbue their messages with social meaning isthrough the creation and use of “emoticons,” “smiley faces,” or “relational icons” createdwith typographic symbols that appear sideways as resembling facial expressions. As early as1982, Hiltz and Turoff stated that “computer conferees also find ways to overcome the lackof personal contact. They have even devised ways of sending computerized screams, hugsand kisses” (cited in Pollack, 1982) by using graphic symbols formed in ASCII characters.Although research clearly indicates that individuals are influenced by the use of nonverbalcues in other contexts (e.g., Childers & Houston, 1984; Shepard, 1967), it is unknown,beyond conjecture, what impact emoticons have in CMC. Do emoticons allow for greaterunderstanding (or misunderstanding) of affect in e-mail messages? Can the absence or pres-ence of emoticons affectively change the meaning of a written message? Also, does the useof graphic nonverbal cues have any effect on the persuasive functions of e-mail communica-tion? This research reports an experiment that attempts to address these questions.

Affect in CMC

CMC in general, and e-mail use in particular, encompasses both impersonal, task-focusedactivities as well as relational development and maintenance activities. The social informa-tion processing (SIP) model (Walther, 1992, 1994) assumes that if communicators in CMChave or expect to have the opportunity to interact over time, they will actively develop socialrelationships no matter what the ostensible purpose of their interaction. The SIP model positsthat users who are unfamiliar with each other form opinions based on textual interactions.This model also suggests that although users achieve “normal” relationships online, doing sorequires a sufficient amount of message exchanges, which, when compared with traditionalFtF communication, take longer to accrue. The lack of nonverbal cues in CMC limits thescope of exchanges, which then require more messages and more time to bring relationaleffects in CMC to the same level as in comparable FtF relationships. A key aspect to the SIPmodel is that users adapt to the medium and find ways to overcome the relative shortage ofcue systems. Whereas the theory originally focused on language content and chronemicadaptations (Walther & Burgoon, 1992; Walther & Tidwell, 1995), Utz (2000) formallyapplied the theory to emoticons, specifying them as another type of accommodation, consis-tent with now-common understandings of emoticon use. For instance, Rezabek andCochenour (1998) asserted that “because the use of e-mail eliminates visual cues such ashead nodding, facial expressions, posture, and eye contact found in face-to-face communica-tion, CMC users often incorporate emoticons as visual cues to augment the meaning of tex-tual electronic messages” (pp. 201-202).

Walther, D’Addario / EMOTICONS 325



Emoticons

The use of the emoticons is well documented in a variety of sources in the CMC literature.Asteroff (1987) referred to emoticons as “relational icons” and documented their use in acase study of e-mail. Sanderson (1993) defined emoticons (or “smileys”) as “a sequence ofordinary characters you can find on your computer keyboard. Smileys are used in e-mail andother forms of communication using computers” (p. 1). Rezabek and Cochenour (1998)defined emoticons as “visual cues formed from ordinary typographical symbols that whenread sideways represent feelings or emotions” (p. 201). Thompson and Foulger (1996)referred to them as “pictographs” and described their use in CMC “to express emotion or assurrogates for nonverbal communication” (p. 226) “suggestive of facial expression . . . [add-ing] a paralinguistic component to a message” (p. 230). Godin (1993) agreed that “until theadvent of the smiley, otherwise known as an emoticon, individuals using electronic commu-nication had no way to indicate the subtle mood changes. They couldn’t tell jokes, use irony,slip in a pun or become bitingly sarcastic” (p. 4). Although this may be an overstatement,Godin argued that when “properly used, a smiley can spice up virtually any form of writtencommunication. Now you can say, ‘Boy, isn’t he intelligent :-)’ and make it quite clear youthink the subject is an idiot.”1

Danet, Ruedenberg-Wright, and Rosenbaum-Tamari (1997) also reflected this assump-tion. Danet et al. defined emoticons as “icons for the expression of emotion, or for markingone’s intent as non-serious. . . . The best known ones are a smile, wink, and frown, respec-tively: :-) ;-) :-(.” Danet and colleagues argued that the development of CMC from awork-related medium to a playful medium is a key ingredient to the continued developmentand use of emoticons (see also Bolter, 1991).

[CMC] refer[red] to a work-related meeting; thus, many people may have expected the generalframe of the messages exchanged to be “serious.” The perception of the medium as cold andanonymous, and lacking in “social presence” because of non-verbal cues such as facial expres-sions also contributed to this expectation. (Danet et al., 1997, n.p.)

Danet et al. suggested that CMC itself has developed into a playful medium.Marvin (1995) also observed that the emoticon is finding its way into everyday CMC

interactions: “Some attention . . . has been lately put on the emergence of ‘smileys’ as a spe-cial feature of writing on the Internet. . . . These symbols are the paralanguage of theinternet.” Marvin discusses CMC within the multiple-user domain, object-oriented environ-ment (MOO). A MOO is a type of computer system that allows for synchronous communica-tion via the Internet. They allow users to “chat” as if they were in the same room and movearound in a virtual space. According to Marvin (n. p.),

The text that may be communicated within MOOs is limited, as it is in most Internet forms, to therange of characters on a typical computer keyboard. . . . An informal, everyday quality is createdthrough the use of smileys, non-standard spelling reflective of vernacular pronunciation, punc-tuation to indicate pauses rather than speech clauses, special symbols borrowed from program-ming languages and an extensive special vocabulary.

Such statements further support the notion that the computer-using society has attemptedto incorporate surrogates for traditional types of interactive nonverbal cues.

Although these reflections about the utility of emoticons seem fairly widespread, less isknown about their actual use and impact. Regarding their use, Witmer and Katzman (1997)

326 SOCIAL SCIENCE COMPUTER REVIEW



examined emoticons in a large sample of Usenet newsgroups postings. In a content analysisof roughly 3,000 messages, they found that 13.2% contained emoticons (or intentional mis-spellings, punctuation, and other CMC-based textual graphics, which the researchers con-tend function in similar fashion). They also found that women (i.e., users using stereotypicalfemale names) used these symbols approximately twice as frequently as men did. If one is toaccept the use of these markers as attempted nonverbal usage, and therefore an attempt at anopen display of affect, these findings support classic structures of gender communication(Quina, Wingard, & Bates, 1987), according to Witmer and Katzman. Wolf (2000) reported asimilar propensity for emoticons being used primarily by women. She also found them usedmore frequently in mostly female online groups than in mostly male groups. In an interestingcross-gender accommodation, Wolf (2000) observed that in mixed-sex discussions, men’suse of emoticons rose to the level of women’s, rather than vice versa. However, women usedemoticons primarily to express humor rather than sarcasm, whereas men used them for sar-casm more than humor. Rezabek and Cochenour (1998) also content-analyzed emoticon useonline, but unlike Witmer and Katzman (1997), they limited their analysis to face representa-tions. In four academic Listservs, emoticons were present in 19.15% to 75% of messages,and in a sample of Usenet groups, 25% of messages. Approximately 53.5% of the face repre-sentations were :-) or :), 10% were ;-), whereas only 7.5% were :-( or :( in the combinedsamples.2

Only two studies of which we are aware have examined impacts of emoticon use. Utz(2000) analyzed their role in relationship development online. Following SIP theory(Walther, 1992), Utz found that Multi-User Dungeon (MUD) players reported using moreemoticons (as well as other MUD-based affective codes) over time, as they learned to accom-modate emotional information to a realm with no nonverbal cues available. She also foundthat the use of such cues was a significant predictor of relationship development in MUDs,accounting for 14% of the variance in users’ frequency of relationship development. Inanother study, Thompson and Foulger (1996) found that an emoticon (presumably a happyface, although the exact emoticon was unspecified) in combination with verbal “flaming”messages modified the perceived hostility of the message (see also Godwin, 1994). Theeffects were inconsistent, however, such that the same emoticon diminished the hostility of amessage showing “tension” but increased the perceived hostility of more antagonistic ver-biage. It is clear that emoticons have become commonplace in CMC, and emoticons haveobviously found their way into the lexicon of the computer-using world. However, research-ers have seldom looked beyond who is using these messages, to document what, if anything,they do conversationally.

Potential Effects

Despite commentators’ descriptions of emoticons as surrogates for nonverbal cues, theyare also, in a literal sense, graphic displays. As such, other contexts of research may informtheir expected impacts. In print advertising—another medium where textual messages areemphasized or accompanied by pictorial components—people were found to rate advertise-ments with a pictorial or graphic component higher than those without; ads with a textual/graphic combination are more effective than those with a textual message alone (Childers &Houston, 1984; Shepard, 1967). Research has also shown that imagery has a positive impacton learning and retention (Lutz & Lutz, 1977). The use of visual cues along with text has alsobeen shown to produce a more positive attitude than text alone (A. A. Mitchell, 1986). Itwould be useful to determine if the nonverbal dialogue in e-mail exerts similar effects oncommunication outcomes as in more traditional FtF or written contexts.




To predict the effects of emoticons in the CMC environment, it may be useful to examineresearch on traditional FtF nonverbal communication. A recent comprehensive review of thenonverbal communication literature by Burgoon, Buller, and Woodall (1996) concluded thatnonverbal behavior predominates the effects of language content in most conditions.Reviewing the “channel reliance” findings, the authors concluded that “if verbal and nonver-bal cues were relatively equal in strength when judged separately, the nonverbal cues domi-nated verbal ones when they were paired together” (p. 137). This tendency depends, how-ever, on the kind of judgment the observer is making about the communication. Whenobservers are asked to assess objective meaning, persuasive communication, or factualdeception, verbal information is more important. For judgments about the speaker’s affect,the reverse is true: “For emotive, relational, or attributional . . . outcomes, nonverbal cuesaccount for varying but greater amounts of the meaning than verbal ones” (Burgoon et al.,1996, p. 140).3

Facial expressions have even greater effects than vocal and spatial nonverbal cues.“Visual cue primacy is also stronger when decoding emotions related to positivity . . . espe-cially when visual cues involve the face,” according to Burgoon et al. (p. 142), who continuethat “the face is particularly important in judging positivity because receivers associate thesmile with positivity, a link that has no analogue in the body and the voice” (p. 142).

According to Burgoon et al. (1996), reliance on nonverbal cues in FtF interaction is evengreater in the case of mixed messages or incongruities between the verbal and nonverbalmessages. “The bulk of evidence . . . would suggest that visual cues tend to be counted morestrongly than vocal cues, which in turn are counted more heavily than verbal ones” (p. 141).This is especially true for facial expressions. Exceptions to this trend may occur by one oftwo means. First, observers may use a “discounting strategy,” that is, they discount or ignorethe valence of one cue if two or more other cues differ from it in an alternative affective direc-tion. Second, observers may be more sensitive to a single cue, even if it conflicts with others,if its valence is extreme. Although these last contingencies make the picture somewhat con-flicted, the overall conclusion seems to be that nonverbal cues—especially visualcues—have impacts as great as, or greater than, verbal messages alone, on the interpretationof emotions in FtF communication and especially in the case of mixed messages. If theseprinciples pertain to the use of emoticons in CMC, similarly strong effects should beexpected.

In a different review, Fridlund, Ekman, and Oster (1987) limited their focus to the impactsof facial expressions, rather than nonverbal cues in general, and their conclusions differ fromBurgoon et al.’s. Initial studies on the impact of facial affect indicated that these expressionsprovide more consensual and accurate judgments of emotion than speech content (or vocalcues). Critics of those studies pointed out, however, that the studies involved contrivedexperimental stimuli. More natural investigations found that speech content was more potentthan visual input, whereas other research found that the relative influence of facial expres-sions to other cues depends on the type of judgment being made about the actor or the actor’sgoal (expression or concealment of emotion). Overall, Fridlund et al. (1987) seem skepticalof strong effects due to facial expressions. They did not discuss mixed messages, as Burgoonet al. (1996) did. Taking this more narrow focus on the face into account, the expected impactof emoticons in CMC is somewhat ambiguous and may be relatively weak.

The CMC literature seems to have overlooked these FtF precedents, or projects strongeffects in spite of them. Although very few studies actually examine emoticon use, almost allcommentators assume that emoticons have meaningful communicative impact. For exam-ple, Rezabek and Cochenour (1998) asserted that “the combination of symbols :-) representsa typical smiley face and conveys the sentiment that the person sending the message [italics




added] and using that particular emoticon is pleased, happy, agreeable or in a similar state ofmind” (p. 201). Even though this assertion may be true, it appears conjectural in Rezabek andCochenour (1998). As no research of which we are aware has demonstrated the actual effectsof emoticons on meaning in electronic communication, it is not known what impacts theyhave on readers of e-mail messages.

Although it is instructive to consider research on the relationship of verbal to nonverbalcues, the use of emoticons is paradoxical with regard to such comparisons. In the following,we first articulate an assumption that limits the expectations for the effects of emoticons andformulate a prediction about emoticon/language combinations that reflects this limitation.Following, reflecting the common assumption that emoticons have impacts similar to thoseof nonverbal cues, we formulate predictions reflecting the expected effects of specific com-binations of emoticons and verbal messages and the ways they may interact.

The Intentionality Issue

Although emoticons may be employed to replicate nonverbal facial expressions, they arenot, literally speaking, nonverbal behavior. This is not a matter of definitional semantics; itaddresses a different issue. Emoticons may or may not have the same intentional connota-tions as physical nonverbal behavior. Whereas nonverbal behavior is thought to reflect gen-eral intentions, many of its forms are perceived by observers to be less controlled and deliber-ate than verbal utterances (see Knapp, Wiemann, & Daly, 1978). Facial expressions areconsidered by some to be among the most controllable of nonverbal cues (Ekman & Friesen,1969). Yet, Birdwhistell (1970) reminds us that people “are not always aware that they are orare not smiling” (p. 33). Furthermore, some forms of facial affect displays—those accompa-nying weeping, for example—are generally considered direct and involuntary representa-tions of internal states (Kendon, 1987). Typed-out textual symbols, whether verbal or iconic,may not be so involuntarily casual, in the minds of receivers. Relative to FtF nonverbal com-munication, emoticons may be considered more deliberate and voluntary. One may uncon-sciously smile FtF, but it is hard to imagine someone typing a :-) with less awareness than ofthe words he or she is selecting. Marvin (1995) recognized this phenomenon in her discus-sion of MOO interaction, stating that

smiles in face-to-face contexts can be strategic or spontaneous and unintentional. In the contextof the MOO . . . every smile must be consciously indicated. In private something flowing acrossthe computer screen might cause a participant to spontaneously smile, but a conscious choicemust be made to type it out; a participant might frown at the keyboard and but [sic] strategicallydecide to type a smile.

However, Marvin does not suggest whether emoticons in CMC differ from facial expressionsin FtF encounters in their impact on receivers. Although it is conceivable that a sender’s gen-eration of emoticons could become habitual and less conscious over time, it is still not clearhow they are interpreted in CMC: as iconic and unconscious like nonverbal facial expres-sions or, like wording, as deliberately encoded elements of intentional communication. If thelatter, whereas physical nonverbal behavior may account for up to 63% of the social meaningin many FtF interactions (Philpott 1983, cited in Burgoon et al., 1996), emoticons may beless potent in this regard. From this perspective, the artificiality and constructed nature ofemoticons suggest that they would have weak effects, if any, given their restricted range andclearly intentional deployment, relative to the connotations of even simple language varia-




tions. This position is reflected in the first hypothesis, in which “valence” refers to the affec-tive direction (i.e., positivity versus negativity) of verbal messages and of emoticons.

Hypothesis 1: Alterations in the valence of verbal messages account for greater variance in theinterpretation of messages than do emoticons.

In other words, the affective dimension of the language in verbal messages makes a biggerdifference than the differences among alternative emoticons do in the way readers interpretthe overall message.

Emoticons as Nonverbal Cues: Pure Messages

Alternatively, emoticons are emblematic of facial expressions, and because they aredescribed as accommodations for the lack of nonverbal cues in CMC, they may have theeffects of nonverbal FtF cues. In relationship with simultaneous verbal behavior, nonverbalbehavior may emphasize, repeat, substitute, or contradict verbal messages (Burgoon, 1985;Ekman & Friesen, 1969), yet CMC commentators discuss emoticons in terms of theiremphatic function or signaling function, not mere repetition or substitution of other-wise-conveyable verbally transmitted meaning. If this is the case, positively valencedemoticons should enhance positively valenced verbal messages, and negative emoticonsmake negatively valenced messages more negative. Thus, the following hypotheses are ten-dered with respect to “pure” combinations of emoticons and verbal messages, that is, mes-sages in which all cues convey either positivity or negativity:

Hypothesis 2a: A smile emoticon, coupled with a positive verbal message, conveys greaterpositivity than a positive verbal message alone.

Hypothesis 2b: A frown emoticon, coupled with a negative verbal message, conveys greaternegativity than a negative verbal message alone.

Mixed Messages

Mixed-message effects are somewhat less predictable. Mixed messages—positive verbalmessages with a negative emoticon or vice versa—may not be readily interpretable. Accord-ing to Leathers (1986), inconsistencies between nonverbal cues and verbal meanings are saidto be ambiguous and problematic for successful communication. If so, the affective meaningof mixed messages should be unclear.

Hypothesis 3a: A smile emoticon, coupled with a negative verbal message, is more ambiguous thana negative “pure message” (a negative verbal message alone or with a frown emoticon) or a posi-tive “pure message” (a positive verbal message alone or with a smile emoticon).

Hypothesis 3b: A frown emoticon, coupled with a positive verbal message, is more ambiguous thana positive “pure message” (a positive verbal message alone or with a smile emoticon) or a nega-tive “pure message” (a negative verbal message alone or with a frown emoticon).

Mixed messages may be more communicationally functional, however, suggesting anintentionally ambiguous or conflicted state. In such a case, the valences of verbal and nonver-bal messages may cancel each other out, consistent with the contradiction relationship ofverbal and nonverbal cues, rendering an overall neutral interpretation. Alternatively, theymay effect an interaction and prompt an interpretation of sarcasm.




Hypothesis 4a: A smile emoticon, coupled with a negative verbal message, conveys less negativitythan a negative “pure message” and less positivity than a positive “pure message.”

Hypothesis 4b: A frown emoticon, coupled with a positive verbal message, conveys less positivitythan a positive “pure message” and less negativity than a negative “pure message.”

Hypothesis 5a: A smile emoticon, coupled with a negative verbal message, conveys greater sar-casm than a negative or positive “pure message.”

Hypothesis 5b: A frown emoticon, coupled with a positive verbal message, conveys greater sar-casm than a positive or negative “pure message.”

Drawing again on known relationships between verbal and nonverbal cues, alternativehypotheses may also be predicted. The relative contributions of verbal and nonverbal com-munication may be combinatory when the two streams are affectively consistent. When theyare dissonant, however, greater decoding weight tends to be placed on the nonverbal features(for a review, see Burgoon et al., 1996), among which facial expressions carry as much astwice the interpretive weight as other cues (i.e., body, then voice; DePaulo & Rosenthal,1979). If e-mail senders are tacitly aware of this combinatory rule, and e-mail readers inter-pret by it, then deployments and interpretations of such combinations may reflect this rule,and the affective valence connoted by the emoticon alone should be interpreted. Such a find-ing would be consistent with a “social meaning” model of nonverbal communication, amodel that dictates that some nonverbal behaviors, regardless of accompanying verbalbehavior or sender characteristics, have consensual meaning (see Burgoon & Le Poire, 1999;Burgoon & Newton, 1991). Further predictions may be offered, refining those previouslyarticulated above, with respect to the magnitude of the effect of the emoticon. Hypotheses 2aand 2b predicted a neutralizing effect of mixed messages. Alternatively, a predominanceeffect for the emoticon can be predicted, such that the emoticon bears the full weight of inter-pretive valence only in the case of mixed messages.

Hypothesis 6a: A smile emoticon, coupled with a negative verbal message, conveys as muchpositivity as a positive “pure message” and more positivity than a negative “pure message.”

Hypothesis 6b: A frown emoticon, coupled with a positive verbal message, conveys as muchnegativity as a negative “pure message” and more negativity than a positive “pure message.”

Winks

A special case may be the emoticon suggesting a smiling wink— ;-) . Although the smil-ing aspect of this symbol suggests positivity, the wink connotes an extra dimension of irony.Among other sources, Rezabek and Cochenour (1998) provided specific connotations forthe wink: “the joke, wry remark, or bit of wit sent over e-mail can be followed by the visualcue ;-) to emphasize the jesting nature of the comment” (p. 202). The wink may have iconicvalue, conducting a social meaning function, that is, no matter the valence of the verbal mes-sage with which it appears, a double meaning is connoted. Whatever the valence of the verbalmessage, the appended wink should diminish or entirely reverse that valence, as well asimply sarcasm.

Hypothesis 7a: A wink emoticon, coupled with a negative verbal message, conveys (a) lessnegativity than a negative “pure message” and (b) greater sarcasm than a negative “pure mes-sage.”

Hypothesis 7b: A wink emoticon, coupled with a positive verbal message, conveys (a) lesspositivity than a positive “pure message” and (b) greater sarcasm than a positive “pure mes-sage.”




Influence

Although when people smile, they often smile about something, that is, some subject ortopic, there is very little research on the role of facial expression in the social influence of atti-tude change, according to Heslin and Patterson (1982). Most of the research focuses on gaze,deception, or on the generation of positive feelings toward the speaker (for a review, seePatterson, 1983) but not toward the topic under discussion. Among the few exceptions, onefocuses on message senders: Mehrabian and Williams (1969) found that communicatorsused more facial activity when instructed to be more persuasive. Another focuses on receiv-ers: Greater smiling is associated with more persuasiveness in a counseling setting(LaCrosse, 1975). Other aspects of gaze do influence others’ various behaviors (such as turntaking), but little else is known about attitude change or its antecedents as far as facial expres-sions are concerned. In research on the effect of graphics on persuasion in CMC, altogetheraside from facial affect, some evidence exists that even simple graphics accompanying a ver-bal message achieve greater social influence than a verbal message alone (King, Dent, &Miles, 1991).

If we assume that the exhibition of a facial expression signals the communicator’s affec-tive attitude toward an object and that social influence should result in the receiver’s attitudetaking the same direction as the sender’s, tentative hypotheses may be posited for facialexpressions in social influence. Thus, for the hypotheses above, the same relationshipsshould be expected to obtain on receivers’ evaluation of the conversational topic, as well astheir interpretation of the communicator’s affect. Measurement of this will provide a repli-cate of the above hypotheses.

Thus, drawing on perspectives about the relationships between verbal and nonverbalcodes, we may expect that emoticons have potent effects, making more extreme the mes-sages with which they are affectively consistent and undermining and/or caricaturing themessages with which they are inconsistent.

METHOD

An experiment was conducted to assess affective and attitude interpretations of emoticoncombinations with verbal phrases. The research comprised a 2 × 4 between-subjects design,with eight stimulus combinations. The two-level variable was verbal message valence: Averbal message was created that was either positive or negative toward the topic it addressed.The four-level variable was emoticons: Each verbal message was combined with either thetypographic symbols :-) , :-( , or ;-) , or as a control condition, with no graphics. Each of thesevariations was embedded in a mock-up of an e-mail message sent from one friend to another,with some material before the critical stimuli and some after them. Each entire message waspresented on a World Wide Web (WWW) page for administration to research participants.

Participants were recruited from two sources at a private technical university in the north-eastern United States. Some participants (n = 160) participated in class, as part of a demon-stration for a large 1st-year course in communication. An additional 66 participants wererecruited from a subject pool for research participation credit in psychology courses. Theseparticipants were recruited to detect whether the first sample showed any bias due to theirstudy of communication and a brief lecture on nonverbal communication. No differenceswere detected between the two subsamples in any subsequent analyses, and the samples arecollapsed in all analyses reported. Participants were removed from the sample if theirresponses were incomplete. The participants’ mean age was 18.48 years (SD = 1.28); 71% of




the sample was male and 29% was female (roughly proportional to the skewed gender ratio atthis university).

The procedures for the presentation and evaluation of stimuli were conducted usingWWW pages and forms (see, e.g., Birnbaum, 2000; Pettit, 1999), as follows. Participantsattended a computer laboratory and were instructed to open a WWW browser to a particularWeb page. The first page explained that the research involved learning about how peopleexperience e-mail messages, and it instructed them that they were about to read an e-mailmessage from one friend to another, after which they would answer a number of questions.Clicking a button activated a computer-based randomization procedure (see Burton &Walther, 2001) that automatically pointed the participants’ browsers to one of the eight pagesdisplaying a variant of the stimulus.

Each message contained a brief discussion about a movie that was showing the followingweekend and then one of two variations of a statement about an Economics course, whichwas one experimental stimulus. The statement about the course was immediately followedby one of the three emoticons or left blank in the case of the control condition. The eight stim-ulus variations are presented in Table 1.

The instructions asked participants to click on a button image on the bottom of that Webpage when they had completed reading the e-mail message. This button linked to the com-mon questionnaire, which also appeared on a WWW page. Participants entered theirresponses via the computers, clicking on “radio buttons” to administer semantic differentialscales or typing in responses such as percentage agreement or ages. Responses were col-lected directly to computer storage, along with automatic indicators of the date and time,using Microsoft FrontPage™ software.

The questionnaire contained the items comprising the dependent variables in thisresearch. Several were created to assess participants’ impressions of the message writer’sattitude toward the topic (the Econ course), the writer’s affect, and other characteristics of themessages. These included several semantic differential items. Four of them asked the partici-pant to assess the e-mail writer’s attitude about the course and the participant’s likelihood totake the course himself or herself (as a measure of social influence). Another five itemsassessed how difficult the message was to understand, how serious the message was, howambiguous the message was, how un/happy the writer was, and how sincere the writer was.Additional items, on a scale from 1 to 100, asked the participants to assess how much happi-ness, sarcasm, urgency, and humor the writer portrayed. The measures are reproduced inTable 2.

In addition, participants’ familiarity with emoticons was assessed in several ways. Partic-ipants were asked to indicate with a yes or no whether they had ever seen or ever used severalcommon emoticons. Furthermore, to see whether they shared consensus on the meaning ofemoticons, we asked participants to match one of the three emoticons used in the study toeach of the following emotions: sad, secretive, happy, honest, sarcastic, seductive, anxious,joking, angry, surprised, disgusted, and afraid. Demographic information was also collected.

RESULTS

Background Data

Participants’ characteristics and experiences with emoticons were analyzed to providecontext for the interpretation of results. Ninety-nine percent of the participants reported hav-ing seen some emoticon in an e-mail message. In contrast to the findings of Witmer and




Katzman (1997) and Wolf (2000), there were no gender differences in the tendency to havesent emoticons in e-mail messages of their own creation: A 1 degree-of-freedom chi-squareof 1.53 was not significant, but too few cases of women who had not sent emoticons (n = 3)undermined this statistic. Descriptively, 84% of men and 92% of women had sent them, for atotal sample proportion of 86%.

No manipulation check was conducted to see whether the participants noticed the specificemoticon presented, as doing so may have drawn artificially great attention to the manipula-tion, and the research was interested in seeing whether emoticons presented effects in as nat-ural a way as possible. Three other checks were performed. The first assessed the degree ofattention participants paid to the e-mail message they read by means of a more innocuouselement, asking them to recall what movie was mentioned in the message. One hundred andtwenty mentioned the correct movie title. Another 7 got part of the title correct, 10 did notremember, and 4 made clear errors. For the most part, there was adequate recall of themessage.

The second check ascertained whether participants were in basic agreement about theconnotations associated with the three emoticons used in the study. Participants had beenasked to select one of the three emoticons to associate with emotion labels. On most emo-


TABLE 1Stimuli

That econ class you asked me about, it’s a joy. I wish all my classes were just like it. :-)That econ class you asked me about, it’s a joy. I wish all my classes were just like it. :-(That econ class you asked me about, it’s a joy. I wish all my classes were just like it. ;-)That econ class you asked me about, it’s a joy. I wish all my classes were just like it.That econ class you asked me about, it’s hell. I wish I never have another class like it. :-)That econ class you asked me about, it’s hell. I wish I never have another class like it. :-(That econ class you asked me about, it’s hell. I wish I never have another class like it. ;-)That econ class you asked me about, it’s hell. I wish I never have another class like it.

TABLE 2Measures

How well does the sender of the message like the econ class? (1 = loathes it, 7 = loves it)Based only on the writer’s recommendation, how likely would you be to take the class yourself? (1 =

would avoid it, 7 = would seek it)Based on the sender’s message, how good a class do you think the econ class is? (1 = great class,

7 = awful class [reverse scored])How does the writer feel about the course? (1 = very negatively, 7 = very positively)On a scale from 1 to 100, with 1 being the least and 100 being the most, how much happiness did the

writer of the message portray?On a scale from 1 to 100, with 1 being the least and 100 being the most, how much sarcasm did the

writer of the message portray?On a scale from 1 to 100, with 1 being the least and 100 being the most, how much urgency did the

writer of the message portray?On a scale from 1 to 100, with 1 being the least and 100 being the most, how much humor did the

writer of the message portray?On the scale below, how easy was it to understand the message? (1 = very difficult, 7 = very easy

[reverse scored])On the scale below, how serious was the message? (1 = not serious at all, 7 = very serious)On the scale below, how happy was the writer of the message? (1= very sad, 7 = very happy)On the scale below, how sincere was the message? (1 = very insincere, 7 = very sincere)How ambiguous is the message? (1 = very ambiguous, 7 = very clear [reverse scored])



tions, there was near consensus. The disagreements were so small in most cases that theydefied statistical analysis. The data for each emotion label, and the emoticons that partici-pants associated with them, are displayed in Table 3. Among those considered most relevantto the present study, there was 98% agreement matching happiness with the ;-) emoticon andsadness with the :-( emoticon. Sarcastic was matched with ;-) in 85% of cases and with :-) in10% of cases. Joking was associated with ;-) most often, by 66.2% of the participants; 32.5%of participants connected joking with :-) but never with :-( . Among these basic connotations,there seems to be a fairly firm semantic understanding of emoticons, whether or not theyfunction this way syntactically. It is noteworthy that the percentages of agreement here arevery similar, if not marginally higher, than the level of agreement people exhibit about realhuman facial expressions of emotion: Ekman and Friesen (1975) reported comparable levelsof agreement for universal facial expressions among U.S. participants, with happiness mostconsensual (97%) and anger the least (67%).

The third check analyzed whether the two versions of the verbal message differed inaffective connotation. Using only the control conditions—verbal messages with noemoticons—a t test compared scores on measures related to happiness (see below). The twomessages invoked significantly different interpretations in the expected directions, t(55) =54.89, p < .001, the positive message with a mean of 4.20 (SD = 1.49, n = 31) and the negativemessage with a mean of 2.73 (SD = 1.30, n = 26), indicating that the manipulation of verbalmessage valence was successful.

Data Transformation

Prior to hypothesis tests, an analysis was conducted to determine whether the affectiveterms presented on the questionnaire represented clusters or independent emotion judg-ments. Items were subjected to a principal components factor analysis with Varimax rota-tion. The analysis produced four factors, accounting for 74% of the variance. Analysisrevealed that the two happiness items formed Factor 1. Factor 2 was composed of the ambi-guity rating and how difficult the message was to understand. Factor 3 contained the itemsfor urgency, sarcasm, and humor. Factor 4 contained seriousness and sincerity assessments(see Table 4).


TABLE 3Association by Percentages of Emoticons With Emotion Labels

Emoticon

:-) or :) ;-) or ;) :-( or :(

Sad 0 2 98Happy 98.3 1.6 0Seductive 8.6 85.4 0.7Anxious 33.1 18.5 38.4Joking 32.5 66.2 0Angry 0 1.3 88.1Secretive 3.3 88.7 0.7Honest 84.8 2.6 3.3Sarcastic 9.9 84.1 0.7Surprised 66.9 17.9 4.6Disgusted 0 1.3 88.1Afraid 0.7 1.3 85.4

NOTE: Percentages that do not add to 100 across rows are due to missing data.



Item groupings were subjected to reliability analyses, with items scaled originally as 1 to100 transformed into sevenths for comparability with seven-interval scales. The items cre-ated to assess the e-mail writer’s attitude about the econ course were also analyzed, andobtained a Cronbach’s alpha reliability of .95; happiness, α = .83; and ambiguity/diffi-culty, α = .74. The urgency/sarcasm/humor composite failed to achieve acceptable reliabil-ity, but removal of the urgency item yielded sarcasm/humor α = .74. Seriousness and sincer-ity did not combine for acceptable reliability and were dropped from further analysis.

Hypothesis Tests

Analyses of variance (ANOVAs) were conducted on each dependent measure for themain and interaction effects of message valence, emoticon, and gender. No effects of genderwere significant, either in interaction with emoticons or message tone, or as a main effect.The gender variable was dropped from further analysis. A similar check was conducted tosee if the naturally occurring differences in e-mail and emoticon experience affected hypoth-esized relationships. Four independent ANOVAs were conducted on all dependent variables,which included the two hypothesized independent factors (verbal message valence andemoticons), plus one of the following additional terms: number of e-mail messages sent perday, number of e-mail messages read per day, whether the participant had ever seen anemoticon in e-mail, and whether the participant had ever used an emoticon in e-mail. None ofthese latter variables produced a significant main or interaction effect, except for experiencehaving sent an emoticon. For this term, there appeared to be a main effect on perceivedurgency and a three-way interaction with emoticon and verbal message on ambiguity. How-ever, inspection of the frequency data revealed that no one who had sent an emoticon himselfor herself had been exposed to the experimental condition featuring a combination of a posi-tive verbal message with the smile emoticon, through randomization. With a mean and vari-ance of zero in this cell, on all analyses, the F test was most likely inflated and unreliable.Thus, experience factors were dropped from further analysis and data were reanalyzed in theplanned 2 × 4 design.


TABLE 4Factor Structure of Emotion Labels

Factor 1 Factor 2 Factor 3 Factor 4

Happy .924 –.006 –.001 .006Sad .921 .001 .001 .006Ambiguous –.004 .900 –.003 .006Difficult to understand .001 .863 –.200 –.002Urgent –.001 .000 .784 –.290Humor .148 –.190 .733 .335Sarcasm –.253 –.215 .659 .412Serious –.310 –.005 .003 –.813Sincere –.330 –.269 .236 .520

Eigenvalues 2.50 2.02 1.17 1.01% variance accounted for 22.26 20.29 18.56 13.39α .83 .74 .74 —

NOTE: Reliability analysis yielded three acceptable composites: happiness, ambiguity, and sar-casm/humor. Items in italics are those retained as measures for the respective composite variables.Otheritems were dropped from further analysis.



TABLE 5Means, Standard Deviations, F Coefficients, and Significant Differences for Message Effects on Interpretations

Message Valence

Negative Negative Negative Negative Positive Positive Positive Positive

Emoticon

:-( None ;-) :-) :-( None ;-) :-)

N 26 26 28 23 32 31 33 22Happiness 2.81a 2.73a 3.23a 3.28a 3.56a,b 4.19b 5.04c 5.31c

1.31 1.30 1.43 1.18 1.52 1.49 1.07 1.11

Fverbiage(1, 213) = 69.90*** Femoticon(3, 213) = 9.22*** Finteraction(3, 213) = 2.42

Course attitude 1.68a 1.56a 1.77a 1.86a 3.70b 4.96c 4.90c 5.18c

0.80 0.64 0.88 1.26 2.10 1.83 1.88 1.78

Fverbiage(1, 213) = 199.23*** Femoticon(3, 213) = 3.37** Finteraction(3, 213) = 2.43*

Ambiguous 5.88b 5.71a,b 5.61a,b 6.00b 4.89a 5.16a,b 5.33a,b 5.05a,b

0.96 0.93 1.21 0.83 1.55 1.49 1.36 1.14

Fverbiage(1, 213) = 16.19*** Femoticon(3, 213) = 0.15 Finteraction(3, 213) = 1.09

Sarcasm/humor 1.35a 1.10a 1.38a 1.58a,b 2.73a,b 1.87a,b 2.43b 2.01a,b

1.27 1.34 1.40 1.53 1.85 2.03 1.81 2.04

Fverbiage(1, 213) = 16.82*** Femoticon(3, 213) = 1.24 Finteraction(3, 213) = 0.74

NOTE: Standard deviations are in italics. Common superscripts indicate homogeneous subsets across rows.*p = .07. **p < .05. ***p < .001. p > .07 not marked.

337

©

2001 SA

GE

Pu

blicatio

ns. A

ll righ

ts reserved. N

ot fo

r com

mercial u

se or u

nau

tho

rized d

istribu

tion

. at P

EN

NS

YLV

AN

IA S

TA

TE

UN

IV on F

ebruary 5, 2008 http://ssc.sagepub.com

Dow

nloaded from


Interaction effects were not significant at the .05 level on any dependent variables in theomnibus 2 × 4 ANOVA, but main effects for verbal message and emoticon did result (and arereported in Table 5). Newman-Keuls analyses were conducted to detect the precise effectsand to identify which combinations of verbal and graphic messages differed from others(also in Table 5). The post hoc analyses suggest that there may be interaction effects, whichappeared to have been suppressed in the ANOVA (because most conditions were not differ-ent from one another, lowering the overall between-conditions variance, a Type II error towhich ANOVA is suceptible; see Rosenthal & Rosnow, 1984). In the following, descriptivepatterns of these differences are explained, including their implications for the hypotheses.

Happiness. For the interpretation of happiness, or positivity, Newman-Keuls analysesrevealed the following: All messages with any negative element, verbal or graphic, wererated significantly more negative than messages with no negative elements. The most nega-tive messages included a positive verbal message accompanied by a frown, which was ratedas unhappy as any combination with a negative verbal message (supporting Hypothesis 6b),but mixed messages did not exhibit the leveling effect posited by Hypothesis 4a or Hypothe-sis 4b. A positive message with no emoticon of any kind was significantly more happy thanthe negative-element messages, but significantly even happier were positive messagesaccompanied by a smile or by a wink, supporting Hypothesis 2a. Overall, a negativity effectwas observed—anything with a negative element in it was less happy—as well as anenhancement effect for pure messages only in the case of positively valenced messages.Among negatively worded combinations, emoticons made no differences. Hypothesis 1 ispartially supported descriptively, in that all negative wordings were less happy than any posi-tive wordings, with one exception.

Course attitude. The course attitude scores provide an important replicate to the happi-ness analysis in assessing positivity/negativity interpretations. As measurement reflects thereceivers’ attitudes about the course, as well as the ostensible message sender’s, it providessome evidence of social influence. That is, the nature of the message affected the partici-pants’ self-reported likelihood of taking the course discussed in the message. Results for rat-ings of attitude toward the course were very slightly different from the happiness ratings andsupported different hypotheses. One aspect was the same: Again, all negative verbal mes-sages, with any emoticon (smile, frown, wink, or no emoticon) were significantly more nega-tive than all other messages. A positive verbal message with a frown had a median interpreta-tion, significantly more positive than the negative messages, but less positive than a positiveverbal message with any other emoticon (smile, wink, or no emoticon), supporting Hypothe-sis 4b. Thus, as far as inferring another’s attitude about a topic and changing one’s own atti-tude, it appears that there is a negativity effect: Any message containing a negative verbal orgraphic element is more negative than any message with no negative elements. Hypothesis 1is indirectly supported: As all negative verbal messages were rated less favorable about thecourse than any positive verbal messages, this prediction stands.

Thus far, the affective analyses—ratings of the messages’ attitude about the course andthe writer’s happiness—do not show many of the expected effects of emoticons. The effectof a happy emoticon enhances the positivity of a positive verbal message, as posited in Hypo-thesis 2a, in the assessment of happiness, but not with regard to the writer’s attitude towardthe topic. Emoticons made no difference in the assessment of happiness for a negativelyworded verbal message, voiding Hypothesis 2b. A frown emoticon affected course attitude,where a frown plus positive statement showed a moderating effect, as is consistent withHypothesis 4b. However, a frown accompanying a positive statement had a different effect




on happiness, where the frown emoticon shifted the interpretation all the way into theunhappy, negative range, as is consistent with Hypothesis 6b. Of the two forms of mixedmessages—a negative verbal message with a positive emoticon or a positive verbal messagewith a negative emoticon—only the latter combination consistently distinguished the mes-sage from the verbal baseline, making it less positive than no emoticon or other emoticons.

Ambiguity. Ambiguity was one of several hypothesized interpretations that might resultfrom mixed messages. Hypotheses 3a and H3b predicted more ambiguity from mixed mes-sages than from pure messages. Ambiguity ratings indicated that although a negative verbalmessage plus smile had the highest raw score on ambiguity, it was not significantly differentthan several other combinations (including positive verbal message plus smile, positive plusnothing, negative verbal plus nothing, and oddly, negative verbiage with a frown), failing tosupport Hypothesis 3a. Moreover, the least ambiguous in absolute scores was the positiveverbal message with a frown, rejecting Hypothesis 3b, although it was not significantly lessambiguous than some other combinations. Statistically, the only differences were that a neg-ative verbal message plus a smile emoticon and a negative verbal message with a frownemoticon were more ambiguous than a positive verbal message with a frown. The biggestdifferences were between two mixed messages. For the positive message plus frown to showsuch little ambiguity suggests that the combination has a more specific connotation, as thenext analysis addresses.

Sarcasm. Ratings on the sarcasm/humor variables were examined to assess Hypotheses 5and 7. Hypotheses 5a and 5b predicted that mixed messages (verbal message valence incon-sistent with emoticon valence) would be seen as more sarcastic than pure messages. Thesehypotheses were not supported. The most sarcastic message, in raw scores, was a positiveverbal message with a wink, but it was not significantly higher than anything except the veryleast sarcastic messages, negative plus frown and negative alone. Although both mixed mes-sages were in the middle, their ratings did not differ from the extremes or from the pure mes-sages specified in the hypotheses from which they were expected to differ.

Winks. Hypotheses 7a and 7b focused on the winking emoticon. A wink emoticon accom-panying a positive verbal message had the highest score on sarcasm/humor, but this ratingdiffered only from the very least sarcastic messages (negative verbal plus nothing and nega-tive verbal plus frown). A wink plus a positive verbal message was not more sarcastic than apositive verbal message with a smile, a frown, or with nothing. The negative verbal messagewith a wink showed no distinguishing ratings at all. To support Hypothesis 7, the winkneeded not only to connote sarcasm but also to affect the positivity/negativity of the verbalmessage with which it appeared. This was not the case either. Despite the wink-plus-positiveverbiage’s high sarcasm rating, the same combination was among the highest on perceivedhappiness and on inferred course attitude. Negative verbiage-plus-wink was no differentfrom negative verbiage with anything on happiness or course attitude. Because of theseresults, Hypothesis 7 is not supported. Although a negative verbal message alone or with afrown seems to be dead serious, winks do not connote greater sarcasm than most other com-binations, and they do not raise sarcasm interpretations relative to the verbal baselines withwhich they are combined. These patterns suggest informal support for Hypothesis 1 only.

Finally, a formal test of Hypothesis 1 was conducted, which had predicted that verbalmessage effects would outweigh emoticon effects or the absence of emoticons. ANOVAs




testing each respective main effect on each dependent variable produced r statistics for eachterm (see Table 6). These r coefficients were transformed to Fischer z� for comparison. Thevariance due to verbal statement was significantly greater than that produced by an emoticonin almost every case. This pertained to course attitude, Z = 7.17, p < .001 (one-tailed); happi-ness, Z = 2.18, p = .015; and message ambiguity, Z = 2.25, p = .012. This difference did notobtain, however, in the case of sarcasm, Z = 1.52, p = .064. It appears that despite some dis-crete variations as noted in the previous results, verbal messages account for the predomi-nance of meaning in e-mail, even when accompanied by emoticons. It is also noteworthy thatthe average variance accounted for by emoticons, an r across conditions of .15 with an R2 =.02, does not approach the meta-analytic estimate of the proportion of social meaning con-veyed by nonverbal cues in FtF communication of about 63% (Philpott, 1983). Even thegreatest effect of emoticons, r = .28, R2 = .078 in the case of happiness interpretations, is stillfar lower. These findings echo Marvin’s (1995) characterization of CMC as creating “worldsmade of words.” The exception to the greater effect of verbal messages—sarcasm—is alsointeresting to note. As is commonly held, it may require a nonverbal behavior—if not anemoticon—to signal sarcasm.

Although these comparisons of verbal communication to emoticon effects are potentiallyinteresting, contemporary nonverbal communication research tends to eschew these grosscomparisons, recognizing that verbal and nonverbal behaviors interact with one another andemanate from the same basic mechanisms. According to Fridlund et al. (1987),

The whole question of how much information is conveyed by separate channels is misleading.There is no evidence that individuals in actual social interaction selectively attend to anotherperson’s face, body, voice, or speech, or that the information conveyed by these channels is sim-ply additive. (p. 190)

From this perspective, and to see if sarcasm indeed was better accounted for by the interac-tion of text and graphic, additional comparisons were conducted between the r coefficientsfor the interaction of verbal-by-emoticon combinations, against emoticon variation alone.These comparisons were significant for all dependent variables, course attitude, Z = 7.56, p <.001; happiness, Z = 3.44, p < .001; ambiguity, Z = 2.32, p = .01; and sarcasm, Z = 1.86, p =.03. These comparisons help clarify that emoticon variations plus verbal message effectshelp communicate sarcasm. On the other hand, emoticon effects alone would not be expectedto be great, or very realistic, when they appear without verbiage. A final analysis revealed nodifferences between verbiage-plus-emoticon effects, compared with verbal message effectsalone, just as we might also expect in some contexts comparing FtF verbal-plus-nonverbalbehavior with verbal behavior alone.

A summary of hypotheses and instances of support are given in Table 7.


TABLE 6Effect Size r for Verbal, Emoticon, and Interaction Effects

Verbal Emoticon Interaction

Course attitude .68 .14 .70Happiness .46 .28 .55Ambiguity .27 .05 .27Sarcasm .27 .13 .30



DISCUSSION

Drawing on the literature regarding the relationships of nonverbal and verbal cues, espe-cially facial affect, the present study formulated numerous hypotheses intended to exploreplausible potential effects of emoticons, on the presumption that they function similarlyonline as nonverbal behavior functions off-line. These hypotheses—often competingones—specified enhancement effects for pure messages, where emoticon and verbal valencewere consistent. Other hypotheses pertained to mixed messages—inconsistent displays ofverbal affect and emoticon valence—predicting diluting effects, contradiction effects, ambi-guity effects, and sarcasm effects. By and large, however, emoticons had few impacts onmessage interpretations as hypothesized, and when they did have an impact, they were notconsistent across replications. For instance, a smile added to a positive verbal messageenhanced happiness ratings but not to an interpretation of the topic to which the statementpertained; a frown added to a positive verbal statement reduced the positivity with regard tofeelings about the topic but not with regard to the writer’s happiness. In most cases,emoticons were overwhelmed by the valence of verbal statements that they accompanied. In


TABLE 7Hypotheses and Support Indications

Hypothesis 1: Alterations in the valence of verbal messages account for greater variance in theinterpretation of messages than do emoticons: Supported.

Hypothesis 2a: A smile emoticon, coupled with a positive verbal message, conveys greater positivitythan a positive verbal message alone: Supported on happiness but not on course attitude.

Hypothesis 2b: A frown emoticon, coupled with a negative verbal message, conveys greater negativitythan a negative verbal message alone: Not supported.

Hypothesis 3a: A smile emoticon, coupled with a negative verbal message, is more ambiguous than anegative “pure message” (a negative verbal message alone or with a frown emoticon) or a positive“pure message” (a positive verbal message alone or with a smile emoticon): Marginal support.

Hypothesis 3b: A frown emoticon, coupled with a positive verbal message, is more ambiguous than apositive “pure message” (a positive verbal message alone or with a smile emoticon) or a negative“pure message” (a negative verbal message alone or with a frown emoticon): Marginally opposite.

Hypothesis 4a: A smile emoticon, coupled with a negative verbal message, conveys less negativitythan a negative “pure message” and less positivity than a positive “pure message”: Not supported.

Hypothesis 4b: A frown emoticon, coupled with a positive verbal message, conveys less positivity thana positive “pure message” and less negativity than a negative “pure message”: Supported oncourse attitude but not on happiness.

Hypothesis 5a: A smile emoticon, coupled with a negative verbal message, conveys greater sarcasmthan a negative or positive “pure message”: Not supported.

Hypothesis 5b: A frown emoticon, coupled with a positive verbal message, conveys greater sarcasmthan a positive or negative “pure message”: Not supported.

Hypothesis 6a: A smile emoticon, coupled with a negative verbal message, conveys as muchpositivity as a positive “pure message” and more positivity than a negative “pure message”: Notsupported (negativity effect).

Hypothesis 6b: A frown emoticon, coupled with a positive verbal message, conveys as muchnegativity as a negative “pure message” and more negativity than a positive “pure message”:Supported on happiness but not on course attitude.

Hypothesis 7a: A wink emoticon, coupled with a negative verbal message, conveys (a) less negativitythan a negative “pure message” and (b) greater sarcasm than a negative “pure message”: Notsupported (negativity effect).

Hypothesis 7b: A wink emoticon, coupled with a positive verbal message, conveys (a) less positivitythan a positive “pure message” and (b) greater sarcasm than a positive “pure message”: Notsupported (negativity effect).



almost all cases, e-mail messages containing emoticons did not generate different interpreta-tions than did messages without emoticons. In terms of the known functional relationships ofnonverbal communication to verbal communication, the emoticon may serve the function ofcomplementing verbal messages at best but not contradicting or enhancing them.

One unpredicted pattern that appeared to persist across all combinations of verbal andgraphic elements was a negativity effect. No matter what the message element—emoticon orwording—when there was a negatively valenced component, the message was seen as morenegative. A pure negative message (i.e., negative statement plus a frown) was not alwaysalone in the group of most negative combinations. Adding a frown to a positive verbal state-ment made it more like one of the combinations with a negative verbal component. Yet, thereverse was not the case: Adding a smile to a negative verbal message did not change the ver-bal connotation (although it may have made the message marginally more ambiguous).Although a negativity effect is not discussed in the nonverbal literature of which we areaware, it is well-known in person-perception dynamics (see Asch, 1946; Kellermann, 1984).Thus, in a communication environment, exclusively verbal and nonphysical, graphic repre-sentations of faces may function more as verbal behavior than as nonverbal behavior func-tions. Alternatively, emoticons may function as “phatic communication” (Malinowski,1923/1946)—ritualized expressions such as “How do you do?” that are exchanged not toconvey meaning but to “fulfill a social function . . . [not] the result of intellectual reflection,nor [to] arouse reflection in the listener” (p. 315). A negative response in this context maydraw attention (e.g., “Hi, how are you?”/“Not very good, actually”) and have an inordinateeffect, redirecting the function from a phatic one to a content exchange.

In each analysis and overall, verbal message content prevailed over the emoticons’ contri-butions. It may be that emoticons are recognized as fleeting, requiring little effort, whereastyping text is slightly more involving and effortful. That which a writer takes the trouble tocompose, in e-mail, might be valued more heavily in an environment in which even banterrequires more work than casual speaking. In FtF interaction, that which is more controllable—language—is considered less revealing than “automatic” nonverbal behavior. In CMC, per-haps, the intentionality of both language and nonverbal surrogates evens things out.Although the SIP theory of social development in CMC (Walther, 1992) contends that CMCusers will use whatever cues they have at their disposal to convey emotion online, it isemphatic that they adapt with alacrity to the presentation of affective information throughverbal means (see also Utz, 2000). The present results seem to support this aspect of the the-ory, from a receiver perspective.

It is a limitation of the present study that each verbal message in the stimuli offered twovalenced phrases, compared to one emoticon. This may violate potential combinatory rulesof verbal/nonverbal communication interpretation: (a) when two or more message elementsare consistent but a third differs, the differing cue may be discounted; (b) when several cuesare similarly valenced, the affective meaning is averaged rather than magnified (Burgoonet al., 1996). However, these are but some among several possibilities of interpretive rules.Nevertheless, future research should adjust stimuli and measurement to be more sensitive tothese factors. Moreover, the present research did not examine the effects of emoticons incombination with affectively neutral or no verbal messages. Doing so might parallel themethod of determining the degree of social meaning produced by nonverbal behavior, inwhich Mehrabian and Ferris (1967) held verbal behavior ambiguous and constant, whilevarying nonverbal affect expressions. That approach led to the oft-quoted statistic that 93%of meaning is communicated nonverbally, which has been criticized as artificial andartifactual. However, it is conceivable that emoticons could be used with neutral verbal mes-




sages, a case in which their meaning might be stronger. Such relationships are currentlyunder study.

Another limitation of the current study is that it did not use actual e-mail messages,directed to the participants as real recipients. Although participants were told to read themessage “as if it were written to you by a friend,” it was indeed not written by a friend. It isvery likely that idiosyncratic interpretations exist within specific relationships, based onfamiliarity with a sender’s style, but the present framework is not amenable to the detectionof such idiosyncracies. Observing others’ messages rather than being an actual recipient mayalso have an effect. Participants reading messages from a true friend may have had highervested interest in the message and thus experience greater cognitive and affective arousal(deTurck & Goldhaber, 1989). Moreover, there are sometimes differences between the per-spectives of participants and observers in judgments about the relational meaning of somenonverbal communication (Burgoon & Le Poire, 1999), although the near unanimity withwhich participants rated emoticons’ meanings might reduce this potential problem (seeBurgoon & Newton, 1991).

Yet another possible reason for these findings is that the emoticon is now overused, andthe impact that it is supposed to have has been diminished, either culturally/historically, or asan individual user is first entertained and later bored with the cuteness of them all. We seemnot to be wearing our smiley-face, “have a nice day” buttons much lately, either. AlthoughRezabek and Cochenour (1998) suggested that the most common, widely recognizedemoticons are most useful for communication, and idiosyncratic ones may not be, that maynot be the case. It may be that emoticons are a tie sign of sorts, signaling common knowledge.Our findings that e-mail or emoticon experience did not affect interpretations, however, con-tradicts this possibility somewhat. Our experimentation with the most common emoticons,and the paucity of effects, might lead researchers to explore less typical examples.

Perhaps emoticons do not actually serve direct socially communicative functions butindirect ones. Although writers of e-mail indeed use emoticons, perhaps the generation of anemoticon acts as a self-signaling cue, prompting the writer to write in such a way that is asexpressive as he or she intends. Such is considered the sometimes function of nonverbal ges-turing, also. Individuals often use gesticulation to help them stimulate their verbal selectionsand regulate their verbiage by using gesture to make a visible representation of that whichthey are trying to articulate as they speak (Freedman, 1977). Emoticons may help the writer,not the reader, in a similar way, by helping to express, to check, and if need be to edit, thatwhich may be unclear during initial message production. As such, emoticons are not com-municative but generative.

In CMC, as in FtF communication, affective displays do not always carry the day. AsGodwin (1994) observed, “ASCII is too intimate”: The impact of language spelled out inplain letters is intense, perhaps more intense in CMC than elsewhere. Yet, even in FtF inter-action, according to Birdwhistell (1970),

Smiles do not override context. That is, insofar as we can ascertain, whatever smiles are andwhatever their genesis, they are not visible transformations of underlying physiological stateswhich are emitted as direct and unmitigated signal forms of that state. (p. 33)

Even more so is the case, at least as interpreted by readers, of the interpretation of smilinggraphics and frowning ones embedded in e-mail messages. Despite widespread familiaritywith these symbols among the computer-using public as well as among scholars and observ-ers, near consensus on their semantic values, and near-universal speculation that they func-




tion as a replacement for the nonverbal FtF cues that CMC eliminates, their actual communi-cative effects are minimal in the context of the language cues they may accompany.

NOTES

1. Popular humorist Dave Barry (1996, pp. 144-146) seems to share this perception, as he explains the criticalimportance of emoticons:

Suppose you’re typing a statement such as:I am feeling happy

The problem with this is that the reader cannot be absolutely, positively, 100 percent sure whatemotion you’re feeling when you type this . . .

I am feeling happy :)See the difference?

Without emoticon:Over 7,000 men died at Gettysburg.

With emoticon:Over 7,000 men died at Gettysburg :(

See the difference? The readers of the second sentence, merely by turning it sideways, will immedi-ately recognize that it is talking about a sad thing.

2. Rezabek and Cochenour (1998) also noted, in one of the Listservs they analyzed, that 33.33% of emoticonswere exhibited by users from the University of Wisconsin–Madison, 21.57% from Texas A&M, 15.69% from SanDiego State, 13.72% from the University of New Mexico, and 17.84% from each of the University of Wyoming andPenn State. Despite descriptions of the Internet as a kind of place-irrelevent cyberspace (e.g., W. J. Mitchell, 1995),apparently regional norms hold. One is reminded of Birdwhistell’s (1970) observation that “smiling varied from onepart of the United States to another” (p. 30).

3. Although these findings are situated in studies of face-to-face interaction, telecommunication research fea-tures conclusions that differ in some respects. When it comes to comparisons between audio, videoconferencing,and other forms of mediated and unmediated interaction, research finds that the voice accounts for a substantialdegree of variance in understanding, whereas the addition of video or face-to-face channels adds very little (e.g.,Ochsman & Chapanis, 1974). However, little in the way of original hypotheses has emerged from these domains tochallenge the findings from face-to-face research (for an exception, see, e.g., Whittaker & O’Conaill, 1997).

REFERENCES

Asch, S. E. (1946). Forming impressions of personality. Journal of Abnormal and Social Psychology, 41,1230-1240.

Asteroff, J. F. (1987). Paralanguage in electronic mail: A case study. Unpublished doctoral dissertation, ColumbiaUniversity, New York.

Barry, D. (1996). Dave Barry in cyberspace. New York: Fawcett Columbine.Birdwhistell, R. L. (1970). Kinesics and context: Understanding body motion. Philadelphia: University of Pennsyl-

vania Press.Birnbaum, M. H. (Ed.). (2000). Psychological experiments on the Internet. San Diego, CA: Academic Press.Bolter, J. D. (1991). Writing space: The computer, hypertext, and the history of writing. Hillsdale, NJ: Lawrence

Erlbaum.Burgoon, J. K. (1985). Nonverbal signals. In M. L. Knapp & G. R. Miller (Eds.), Handbook of interpersonal com-

munication (pp. 344-392). Beverly Hills, CA: Sage.Burgoon, J. K., Buller, D. B., & Woodall, W. G. (1996). Nonverbal communication: The unspoken dialogue (2nd

ed.). New York: McGraw-Hill.Burgoon, J. K., & Le Poire, B. A. (1999). Nonverbal cues and interpersonal judgments: Participant and observer

perceptions of intimacy, dominance, composure, and formality. Communication Monographs, 66, 105-124.Burgoon, J. K., & Newton, D. A. (1991). Applying a social meaning model to relational message interpretations of

conversational involvement: Comparing observer and participant perspectives. Southern Communication Jour-nal, 56, 96-113.

Burton, M. C., & Walther, J. B. (2001, January). The value of Web log data in use-based Web design and testing.Journal of Computer-Mediated Communication, 6(2). http://www.ascusc.org/jcmc/vol6/issue3/burton.html




Carey, J. (1980). Paralanguage in computer mediated communication. In N. K. Sondheimer (Ed.), The 18th annualmeeting of the Association for Computational Linguistics and parasession on topics in interactive discourse:Proceedings of the conference (pp. 67-69). Philadelphia: University of Pennsylvania.

Childers, T. L., & Houston, M. J. (1984). Conditions for a picture-superiority effect on consumer memory. Journalof Consumer Research, 11, 643-654.

Danet, B., Ruedenberg-Wright, L., & Rosenbaum-Tamari, Y. (1997). “HMMM . . . WHERE’S THAT SMOKECOMING FROM?” Writing, play and performance on Internet Relay Chat. Journal of Computer-MediatedCommunication, 2 (4). Retrieved June 15, 2000, from the World Wide Web: http:www.ascusc.org/jcmc/vol2/issue4/danet.html

DePaulo, B. M., & Rosenthal, R. (1979). Ambivalence, discrepancy, and deception in nonverbal communication. InR. Rosenthal (Ed.), Skill in nonverbal communication: Individual differences (pp. 204-248). Cambridge, MA:Oelgeschlager, Gunn, & Hain.

deTurck, M. A., & Goldhaber, G. M. (1989). Effectiveness of product warning labels: Effects of consumer’s infor-mation processing objectives. Journal of Consumer Affairs, 23(1), 111-126.

Ekman, P., & Friesen, W. V. (1969). The repertoire of nonverbal behavior: Categories, origins, usage, and coding.Semiotica, 1, 49-98.

Ekman, P., & Friesen, W. V. (1975). Unmasking the face: A guide to recognizing emotions from facial cues.Englewood Cliffs, NJ: Prentice Hall.

Freedman, N. (1977). Hands, words and mind: On the structuralization of body movements during discourse and thecapacity for verbal representation. In N. Freedman & S. Grand (Eds.), Communicative structures and psychicstructures: A psychoanalytic interpretation of communication (pp. 109-132). New York: Plenum.

Fridlund, A. J., Ekman, P., & Oster, H. (1987). Facial expression of emotion: Review of the literature, 1970-1983. InA. W. Siegman & S. Feldstein (Eds.), Nonverbal behavior and communication (2nd ed., pp. 143-244). Hillsdale,NJ: Lawrence Erlbaum.

Godin, S. (1993). The smiley dictionary. Berkeley, CA: Peachpit.Godwin, M. (1994, April). ASCII is too intimate. Wired, pp. 69-70.Heslin, R., & Patterson, M. L. (1982). Nonverbal behavior and social psychology. New York: Plenum.Kellermann, K. (1984). The negativity effect and its implications for initial interaction. Communication Mono-

graphs, 51, 19-56.Kendon, A. (1987). On gesture: Its complementary relationship with speech. In A. W. Siegman & S. Feldstein

(Eds.), Nonverbal communication and behavior (pp. 65-98). Hillsdale, NJ: Lawrence Erlbaum.Kiesler, S., Siegel, J., & McGuire, T. W. (1984). Social psychological aspects of computer-mediated communica-

tion. American Psychologist, 39, 1123-1134.King, W. C., Dent, M. M., & Miles, E. W. (1991). The persuasive effect of graphics in computer-mediated communi-

cation. Computers in Human Behavior, 7, 269-279.Knapp, M. L., Wiemann, J. M., & Daly, J. A. (1978). Nonverbal communication: Issues and appraisal. Human Com-

munication Research, 4, 271-280.Kraut, R., Mukhopadhyay, T., Szczypula, J., Kiesler, S., & Scherlis, B. (1999). Information and communication:

Alternative uses of the Internet in households. Information Systems Research, 10, 287-303.LaCrosse, M. B. (1975). Nonverbal behavior and perceived counselor attractiveness and persuasiveness. Journal of

Counseling Psychology, 22, 563-566.Leathers, D. G. (1986). Successful nonverbal communication: Principles and applications. New York: Macmillan.Lutz, K. A., & Lutz, R. J. (1977), Effects of interactive imagery on learning: Applications to advertising. Journal of

Applied Pyschology, 62, 493-498.Malinowski, B. (1946). The problem of meaning in primitive languages. Supplement 1 in C. K. Ogden & I. A. Rich-

ards, The meaning of meaning (8th ed., pp. 296-336). New York: Harcourt, Brace, & Co. (Original work pub-lished 1923)

Marvin, L. E. (1995). Spoof, spam, lurk and lag: The aesthetics of text-based virtual realities. Journal of Com-puter-Mediated Communication, 1(2). Retrieved June 15, 2000, from the World Wide Web:http://207.201.161.120/jcmc/vol1/issue2/marvin.html

McCormick, N. B., & McCormick, J. W. (1992). Computer friends and foes: Content of undergraduates’ electronicmail. Computers in Human Behavior, 8, 379-405.

Mehrabian, A., & Ferris, S. R. (1967). Inference of attitudes from nonverbal communication in two channels. Jour-nal of Consulting Psychology, 31, 248-252.

Mehrabian, A., & Williams, M. (1969). Nonverbal concomitants of perceived and intended persuasiveness. Journalof Personality and Social Psychology, 13, 37-58.

Mitchell, A. A. (1986). The effect of verbal and visual components of advertisements on brand attitudes and attitudetoward the ad. Journal of Consumer Research, 13, 12-24.




Mitchell, W. J. (1995). City of bits: Space, place, and the Infobahn. Boston: MIT.Negroponte, N. (1995). Being digital. New York: Alfred A. Knopf.Ochsman, R., & Chapanis, A. (1974). The effects of 10 communication modes on the behavior of teams during

co-operative problem-solving. International Journal of Man-Machine Studies, 6, 579-619.Patterson, M. L. (1983). Nonverbal behavior: A functional perspective. New York: Springer-Verlag.Pettit, F. A. (1999). Exploring the use of the World Wide Web as a psychology data collection tool. Computers in

Human Behavior, 15, 67-71.Pew Internet & American Life Project. (2000). Tracking online life: How women use the Internet to cultivate rela-

tionships with family and friends. Retrieved May 16, 2000, from the World Wide Web: http://www.pewinternet.org/reports/pdfs/Report1.pdf

Philpott, J. S. (1983). The relative contribution to meaning of verbal and nonverbal channels of communication: Ameta-analysis. Unpublished master’s thesis, University of Nebraska, Lincoln.

Pollack, A. (1982, May 27). Technology: Conference by computer. New York Times, p. D2.Quina, K., Wingard, J. A., & Bates, H. G. (1987). Language style and gender stereotypes in person perception. Psy-

chology of Women Quarterly, 5(2), 111-122.Rezabek, L. L., & Cochenour, J. J. (1998). Visual cues in computer-mediated communication: Supplementing text

with emoticons. Journal of Visual Literacy, 18, 201-215.Rice, R. E., & Love, G. (1987). Electronic emotion: Socioemotional content in a computer-mediated network. Com-

munication Research, 14, 85-108.Rosenthal, R., & Rosnow, R. L. (1984). Essentials of behavioral research: Methods and data analysis. New York:

McGraw-Hill.Sanderson, D. (1993). Smileys. Sebastopol, CA: O’Reilly.Sarbaugh-Thompson, M., & Feldman, M. S. (1998). Electronic mail and organizational communication: Does say-

ing “hi” really matter? Organization Science, 9, 685-698.Shepard, R. N. (1967). Recognition memory for words, sentences, and pictures. Journal of Verbal Learning and

Verbal Behavior, 6, 156-163.Thompson, P. A., & Foulger, D. A. (1996). Effects of pictographs and quoting on flaming in electronic mail. Com-

puters in Human Behavior, 12, 225-243.Tidwell, L. C., & Walther, J. B. (2000, July). Getting to know one another a bit at a time: Computer-mediated com-

munication effects on disclosure, impressions, and interpersonal evaluations. Paper presented at the Interna-tional Conference on Language and Social Psychology, Cardiff, Wales.

Utz, S. (2000). Social information processing in MUDs: The development of friendships in virtual worlds. Journalof Online Behavior, 1(1). Retrieved June 15, 2000, from the World Wide Web: http://www.behavior.net/JOB/v1n1/utz.html

Walther, J. B. (1992). Interpersonal effects in computer-mediated communication: A relational perspective. Com-munication Research, 19, 52-90.

Walther, J. B. (1994). Anticipated ongoing interaction versus channel effects on relational communication in com-puter-mediated interaction. Human Communication Research, 20, 473-501.

Walther, J. B. (1997). Group and interpersonal effects in international computer-mediated collaboration. HumanCommunication Research, 23, 342-369.

Walther, J. B., & Burgoon, J. K. (1992). Relational communication in computer-mediated interaction. Human Com-munication Research, 19, 50-88.

Walther, J. B., & Tidwell, L. C. (1995). Nonverbal cues in computer-mediated communication, and the effect ofchronemics on relational communication. Journal of Organizational Computing, 5, 355-378.

Whittaker, S., & O’Conaill, B. (1997). The role of vision on face-to-face and mediated communication. InK. E. Finn, A. J. Sellen, & S. B. Wilbur (Eds.), Video-mediated communication (pp. 23-49). Mahwah, NJ: Law-rence Erlbaum.

Witmer, D., & Katzman, S. (1997). On-line smiles: Does gender make a difference in the use of graphic accents?Journal of Computer-Mediated Communication, 2(4). Retrieved May 23, 2000, from the World Wide Web:http://www.ascusc.org/jcmc/vol2/issue4/witmer1.html

Wolf, A. (2000). Emotional expression online: Gender differences in emoticon use. CyberPsychology & Behavior,3, 827-833.

Joseph B. Walther (Ph.D., University of Arizona) is an associate professor with appointments in communi-cation, social psychology, and information technology at Rensselaer Polytechnic Institute in Troy, New York.




His research involves the relational dynamics of computer-mediated communication in groups, organiza-tions, educational settings, and personal relationships. He can be reached at 4202 Sage Lab, Dept. of Lan-guage, Literature, & Communication, R.P.I., Troy, NY 12180; e-mail: [email protected].

Kyle P. D’Addario (B.A., State University of New York at Buffalo) is a doctoral student in communicationand rhetoric at Rensselaer Polytechnic Institute in Troy, New York.




The Impact of the Emoticons Walter and d'Addario

Documents

Transcript of The Impact of the Emoticons Walter and d'Addario